Subject: Re: ancillary data alignment and binary backward compatibility
To: Jun-ichiro itojun Hagino <firstname.lastname@example.org>
From: Chris G. Demetriou <email@example.com>
Date: 03/07/2000 23:16:50
Jun-ichiro itojun Hagino <firstname.lastname@example.org> writes:
> >Why is this a good thing to have a sysctl for?
> >* it's not gonna change on a given MACHINE_ARCH. (if it does, well,
> >something _else_ is seriously wrong.)
> >* it's easier, nicer, faster code is produced if you do
> >masks/shifts/ands etc. with constants known at compile time than if
> >you do them all as needed at run-time with variables.
> ALIGNBYTES can be different between the kernel and the userland,
> if you run sparc32 binary on sparc64 kernel.
> (and bunch of emails near this one)
I read that bit. That's not a justification.
For binaries of MACHINE_ARCH sparc (i.e. sparc32 binaries), ALIGNBYTES
(and friends) have a particular definition.
Emulation of the binaries for that different MACHINE_ARCH (and yes, it
is emulation, just as linux emulation on x86 is emulation; it's a
different ABI) must respect that definition, to be correct.
The compiler, for instance, isn't gonna start using a larger structure
alignment for random structures compiled for sparc32 when run on
sparc64 -- it doesn't, can't, and shouldn't know about what the kernel
thinks the alignment "should be."
Chris Demetriou - email@example.com - http://www.netbsd.org/People/Pages/cgd.html
Disclaimer: Not speaking for NetBSD, just expressing my own opinion.